« All work
[ Case study ] — Call centers

An AI phone agent that plugs into a call center’s own dialer

Plugs into a call center’s existing dialer, qualifies leads in a natural voice and hands interested ones to a human — replying in under a second.

AI calling systemOwn product · piloting with an outbound call centerVoice AI + telephonyPilot-ready2026
Live calls · 12 active
AI calling system illustration of the productPrivate system — shown as an illustration
ClientOwn product · piloting with an outbound call center
IndustryCall centers
ProductVoice AI + telephony
My roleFounder & solution architect
[ 01 / 07 ] — About the client

Who it’s for

Outbound call centers run lead-generation campaigns — insurance, solar, Medicare — with rows of agents repeating the same qualifying script thousands of times a day.

[ 02 / 07 ] — The challenge

What needed to change

Headcount and turnover cap how many calls a center can make. Cold leads hang up the moment something sounds robotic, so reply speed and handling interruptions matter enormously. And most call centers won’t give up the dialer they already use.

[ 03 / 07 ] — The build

What I built

  1. [01]
    An AI agent that follows a campaign script on a normal phone line — asks the qualifying questions, handles objections, never quotes a price
  2. [02]
    A plug-in for the center’s existing dialer: their dialer places the call, the AI answers it — with any phone carrier
  3. [03]
    Warm transfer of qualified, willing leads to a live human agent in the middle of the call
  4. [04]
    Natural turn-taking: the caller can interrupt, and the AI checks in if the line goes quiet
  5. [05]
    Pre-recorded script lines for the fixed parts of the script, which roughly halves voice costs; compliance disclaimers read word for word
  6. [06]
    A record of every call: recording, transcript, reply times, cost and outcome
  7. [07]
    Results pushed to the client’s CRM or dialer automatically, with retries
  8. [08]
    Do-not-call suppression, per-client line limits and a multi-client console with reports
[ 04 / 07 ] — How it works

How it works, step by step

  1. [01]

    The center’s dialer calls a lead

  2. [02]

    Live audio streams to the AI agent

  3. [03]

    It listens and replies from the campaign script

  4. [04]

    Qualified leads are warm-transferred to a human

  5. [05]

    The outcome is posted back to the client’s system

[ 05 / 07 ] — Results

Results so far

0.8 s

median AI reply time on real phone calls

10

simultaneous calls in load testing — capacity planned for 50+

$0.0135

AI cost per talk-minute, measured on real calls

[ 06 / 07 ] — Key decisions

Decisions that mattered

  1. [01]
    Keep the client’s dialer — “their dialer dials, our AI answers” — so there’s nothing new to adopt.
  2. [02]
    Tap the audio at the phone switch rather than through one carrier, so any carrier works.
  3. [03]
    Pre-record the repeated lines to cut voice costs, and keep listening tuned for speed.
[ 07 / 07 ] — Under the hood

The technology

VICIdial / AsteriskPython + FastAPI (asyncio)Real-time speech-to-text, LLM and text-to-speechSupabaseReact consoleDocker on Hetzner

For the technically curious. You don’t need to know any of this to work with me.

[ Get started ] — Book a call

Have something to build?

Tell me what’s slowing your business down. I’ll tell you honestly what software can fix — and what it can’t.

[01]Project

Product build

From idea to launched product. Clear scope, weekly releases.

[02]Ongoing

Ongoing partner

Your part-time technical lead: new features, fixes and advice every month.

[03]Add-on

Add AI to your business

An AI assistant, automation or private AI added to what you already run.